How to convert yolox to tflite int8?

[Alter-Y]

I transformed the weights of yolox from torch to tensorflow keras, and now i want to transform keras model to tflite. But when using int8 quantization, the output array of confidence and class ([1, 8500, 5:]) are all 0. I found it's because yolox performed convolution on box, confidence, and class, respectively, in yolo_head.py. And box outputs are not activated with sigmoid, it causes problem for concatenation layers when the input ranges have a great difference. So, how to fix this problem?